Calculate goodness of fit (r2_score) based on posterior expectation rather than posterior prediction

[drbenvincent]

• Previously, the r2_score was calculated by comparing y_true and y_pred, where y_pred was the likelihood-influenced predictions i.e. with the observation noise added. Now we set y_pred to be based on the model's expected outcomes.
• Closes Fix failing doctest #428

---

📚 Documentation preview 📚: https://causalpy--429.org.readthedocs.build/en/429/

[drbenvincent]

Oh, I think this is because of rng being used multiple times (in the docstring there was a call to rng.normal before the model.score) which changed its state. Simple fix will be to rearrange order of calculations in the docstring. Though I'm not totally on top of why doctests were passing locally and not remotely.

[drbenvincent]

Why are these f*king doctests passing locally but not remotely. Does rng have different behaviour on different systems or numpy versions?

[drbenvincent]

These doctests have been a constant source of frustration. So far I've kept them around in the hope that they reveal errors or misunderstandings. But if it's nonsense like different behaviours of rng on different OS's or numpy versions, then we'll at the least get rid of output checking in the doctests.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 94.40%. Comparing base (142694f) to head (cdf2ee5).
Report is 4 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #429   +/-   ##
=======================================
  Coverage   94.40%   94.40%           
=======================================
  Files          31       31           
  Lines        1985     1985           
=======================================
  Hits         1874     1874           
  Misses        111      111           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests